Contents
The Privacy Protection Authority has published a draft guideline that sets out a simple position: an artificial intelligence system that processes personal data is subject to the Privacy Protection Law, exactly like any other database. Amendment 13 to Israel's Privacy Protection Law gave the Authority expanded enforcement powers, including administrative fines.
The practical conclusion: there is no "gray area" in which entering information into a chatbot does not count as processing. If you entered a customer's name — you processed personal data.
The First Question: What Counts as Personal Data at All
This is where most organizations get it wrong. Personal data is not only an ID number and a full name. It is any detail that makes it possible to identify a person, including in combination.
- A service call transcript — even without a name, if it contains a customer number or incident details.
- An email pasted into a chatbot "just to draft a reply" — including the signature and the address.
- Résumés entered for initial screening — clearly personal data.
- A contact list pasted in "to arrange it in a table".
- A sales report with customer names, even if the goal is only trend analysis.
Five Steps, in Order
Map what is already happening
Before any policy — find out what is in use today. A short anonymous survey gives a more accurate picture than any monitoring tool, because people answer honestly when there is no sanction.
Separate personal accounts from the enterprise version
This is the most significant distinction. Enterprise versions come with an agreement, a commitment that data is not used for training, and controls. A personal free-tier account has none of that — and with some vendors the content is used to improve the model unless the user has actively opted out.
Define what may be entered — as a list, not a principle
"Don't enter sensitive information" is an instruction nobody knows how to apply. "Don't enter customer names, account numbers, medical information or source code" is an instruction people can act on.
Establish the consent it relies on
Information collected years ago, for a different purpose, is not necessarily covered by the consent given at the time. Before entering historical customer data into an AI system, this question must be answered.
Document
Which tools were approved, who approved them, when, and what the instruction is. In an audit, an organization without documentation is in the same position as an organization without a policy.
The Common Mistake: Banning It
A blanket ban looks like the safe solution, and it is almost always the worst one. It does not stop the use — it moves it to personal phones, where you have no visibility at all.
The approach that works: provide an approved tool that is good enough, and explain why. An employee who has a proper tool does not go looking for another.
Amendment 13 requires certain organizations to appoint a Data Protection Officer (DPO). If you have not yet checked whether you fall within its scope — that is the first question, before any discussion of AI.
What This Guide Does Not Cover
- This guide is not legal advice. Whether Amendment 13 applies to a specific organization depends on the size of its databases and the type of information.
- We do not cover sector-specific requirements here — healthcare, finance and defense are also subject to their own sector regulator.
- The Authority's guideline was published as a draft. The final wording may change.
Sources
Want to Go Through It Together?
A guide is a starting point. An assessment of your environment will give you a precise answer.
